The colab doesn’t allow me GPU T4 option, just GPU.
Is it reasonable that each epoch takes about 18 minutes?
I tried different batch sizes.
The colab doesn’t allow me GPU T4 option, just GPU.
Is it reasonable that each epoch takes about 18 minutes?
I tried different batch sizes.
It could be be, depending on the dataset size!
This sepcific exercise is with dogs and cats full dataset, 25k images, with augmentation. Input size 150*150.
The dataset is big thats for sure!
Thanks.
The reason for slowness is ImageDataGenerator
(see this). Here’s my recommendation:
image_dataset_from_directory
to iterate on model architecture.image_dataset_from_directory
with ImageDataGenerator
after finalizing the model architecture and pay the price once to please the grader.